ELUVIXSovereign AI Router

Canvas Layouts & Topology Templates

Eluvix includes a robust Layout Management System that allows developers to save, organize, export, and switch between complete routing topologies. It ships with 12 pre-configured, production-grade architectural presets designed for enterprise security, local-first engineering, autonomous coding, and high-throughput evaluation.

Layout Management System

In the Visual Routing Editor (/editor), the top toolbar provides direct access to layout management tools:

  • Browse Presets & Layouts (LayoutBrowserModal): Open the interactive catalog of available topologies, filter by category (starter, privacy, performance, agent, custom), or search by tags.
  • Save As (SaveAsModal): Save your current canvas state under a unique slug, name, category, and descriptive tags.
  • Set as Active: Marking a layout active instructs the backend Switchyard engine to immediately compile and hot-reload that configuration for incoming API traffic.
  • Restore Defaults: One-click reset that safely repopulates the 12 official starter templates without wiping custom user layouts.

Storage & Persistence

Layout definitions are stored in the XDG user data directory (~/.local/share/eluvix/layouts/ or ./data/layouts/) as structured JSON files containing node coordinates, handle connection edges, viewport zoom levels, and custom node properties.

The 12 Built-In Topology Templates

1. Full Production Stack (default)

  • Category: starter
  • Focus: Enterprise All-Rounder
  • Architecture: 
    [ Client Input ] ──► [ Privacy Guard ] ──► [ Semantic Guard ]
                                                    │
                                                    ▼
[ Cloud Fallback ] ◄── [ Local Engines ] ◄── [ Switchyard Balancer ]
  • Description: The recommended out-of-the-box configuration. Combines PII anonymization, DeBERTa prompt injection defense, Switchyard cost-complexity balancing, local Lemonade/Ollama nodes, and cloud API fallbacks.

2. 100% Local Airgapped Cluster (local-airgapped)

  • Category: privacy
  • Focus: Sovereign Offline Inference
  • Architecture: Enforces zero outbound internet access. Routes exclusively to local Lemonade, vLLM, and Ollama instances.
  • Best For: Defense, healthcare, classified enclaves, or offline workstation deployments where data leakage to commercial cloud APIs is strictly prohibited.

3. Local-First Coding with AGY Frontier (local-coding-agy)

  • Category: agent
  • Focus: Software Engineering & Code Generation
  • Architecture: Dispatches routine code completions and unit tests to local fast coding models (e.g. Qwen 2.5 Coder 7B), while automatically code-escalating architectural design and complex refactoring tasks to the Google Antigravity CLI (agy) with extended thinking budgets.

4. Ultra-Low Latency Edge Chat (low-latency-edge)

  • Category: performance
  • Focus: Interactive Real-Time Applications
  • Architecture: Configured with strategy: "lowest-latency". Measures real-time round-trip latency in SQLite telemetry logs and dispatches queries to whichever engine delivers sub-second response times.

5. Tiered Cost Cascade Routing (tiered-cost-cascade)

  • Category: performance
  • Focus: Cloud Bill Reduction
  • Architecture: Configured with strategy: "cascade-cost". Always executes against $0 local hardware first. If local nodes hit capacity, timeout, or report OOM, it escalates to mid-tier cloud APIs before touching expensive frontier models.

6. Zero-Data-Leak Airgap Shield (zero-data-leak-shield)

  • Category: privacy
  • Focus: Hardened Compliance & Threat Defense
  • Architecture: Chains Privacy Guard (cryptographic SHA-256 hash masking) with Semantic Guard (strict action: "block" threshold at 0.90) and an isolated Quarantine Sink. Banned terms and suspicious prompt structures are blocked with zero information disclosed.

7. Multi-Agent Autonomous Engineering (multi-agent-orchestrator)

  • Category: agent
  • Focus: Complex Autonomous Workflows
  • Architecture: Features a central Task Router projecting multiple intent ports:
    • #intent-code: Routed to OpenAI Codex CLI
    • #intent-reasoning: Routed to Google Antigravity AGY
    • #intent-refactor: Routed to Anthropic Claude Code CLI
    • #intent-default: Routed to Switchyard fallback

8. Direct Google Antigravity AGY (direct-antigravity)

  • Category: agent
  • Focus: Dedicated Gemini Agent Loop
  • Architecture: Connects the entrypoint directly to the Antigravity Node, configured with gemini-3.8-flash, planning execution mode, autonomous shell permissions, and extended thinking budgets up to 4,096 tokens.

9. Round-Robin Model Evaluation Lab (model-eval-lab)

  • Category: starter
  • Focus: A/B Benchmarking & Output Comparison
  • Architecture: Configured with strategy: "round-robin". Distributes identical prompt workloads evenly across multiple competing local and cloud models to benchmark latency, token generation, and response quality.

10. Multimodal & Vision Pipeline (multimodal-vision-pipeline)

  • Category: performance
  • Focus: Mixed Image & Text Ingestion
  • Architecture: Automatically filters candidate upstreams to include only engines advertising the vision capability tag whenever image payloads are detected, gracefully routing text-only queries to standard low-memory models.

11. MCP Agent Toolkit Hub (mcp-agent-hub)

  • Category: agent
  • Focus: Tool Use & External API Integration
  • Architecture: Connects the Agent Gateway to multiple MCP Tool Nodes (Filesystem, WebScraper, Database Inspector). Configured with human-in-the-loop confirmation (mode: "confirm") and automated output PII sanitization.

12. Deep Observability & Semantic Debug (deep-observability-debug)

  • Category: performance
  • Focus: Forensic Analysis & Compliance Auditing
  • Architecture: Couples the Observability Node at 100% trace sampling rate with OpenTelemetry collector exports, Langfuse graph logging, and SLA alert boundaries set to 2,000ms.

Custom Layout CRUD via REST API

In addition to the visual canvas, layouts can be managed programmatically via /api/v1/layouts:

# List all saved layouts and current active slug
curl -H "Authorization: Bearer <ADMIN_PASSWORD>" https://127.0.0.1:3000/api/v1/layouts

# Activate a specific layout
curl -X POST https://127.0.0.1:3000/api/v1/layouts/active \
  -H "Authorization: Bearer <ADMIN_PASSWORD>" \
  -H "Content-Type: application/json" \
  -d '{"slug": "local-coding-agy"}'

# Restore official example templates
curl -X POST https://127.0.0.1:3000/api/v1/layouts/restore-examples \
  -H "Authorization: Bearer <ADMIN_PASSWORD>"